The AI Inference Economy – What is Edge AI, Exactly?

This article is part of a series exploring two complementary investment themes. The ROBO Global Artificial Intelligence Index (THNQ) captures the digital AI ecosystem, including AI-semiconductors, cloud infrastructure, cybersecurity, connectivity, and applications. The ROBO Global Robotics and Automation Index (ROBO) captures the physical automation ecosystem, including robotics, sensors, semiconductors, and industrial systems. Together, they tell the story of autonomy as intelligence shifts from the cloud to the edge.

At the end of the day, the future robotic and AI-augmented economy will transform every industry. Underneath it all, the drivers are physics, economics, and trust. 

What makes edge AI compelling is that demand is broadly distributed. It is not concentrated in a handful of hyperscale buyers. For example, NVIDIA disclosed that two direct customers represented roughly 39 percent of total revenue in their most recent earnings report, with more than half of data center revenue coming from only three customers. Edge demand is more diverse, global, and underappreciated by markets.

Data centers still matter. They power training, simulation, and a growing share of inference. But the next leg of growth will live at the edge.

First principles: What is inference? What is “the edge”?

Inference is AI in use. You ask a model a question. It senses inputs and analyzes them, then acts by returning an output. Robotics uses the same loop: sense, analyze, act. 

Today, much of that loop runs in the cloud. As models get faster, cheaper, and smaller, the loop moves closer to where data is created: on your phone, a camera, a robot, a vehicle, or a factory tool. Currently, the majority of inference exists as the query from your fingertips on the Gemini, Claude, Grok, or ChatGPT interface on your phone transmitting to the data center and back with a response. It won’t always be this way. 

Edge AI is the physical nexus with the real world. It runs in real time, often on tight power and size budgets.

Real-World Applications of Edge AI

Edge AI is already shaping daily life. It enables real-time decision making where speed, privacy, or connectivity matter. Instead of sending every request back to a distant data center, intelligence sits inside the device, machine, or vehicle itself. This shift allows faster responses, lower energy use, and new applications that weren’t possible before.

Examples:

• Autonomous vehicles: Cars and trucks use on-board GPUs and sensors to process lidar, radar, and camera data in real time. They must detect pedestrians, cyclists, and other vehicles in milliseconds, even without a network connection.
• Drones: Consumer and military drones use vision processors for navigation, obstacle avoidance, and mapping. They can also share “cooperative perception” data with other drones or ground systems to extend what each one can see.
• Robotics: Factory robots detect defects in products on assembly lines and adjust movements on the fly. Service robots in warehouses or hospitals navigate safely in human environments without relying on cloud latency.
• Healthcare: Wearables and diagnostic devices analyze data locally to protect patient privacy. Examples include portable ultrasound machines with embedded AI or glucose monitors that interpret results instantly.
• Retail and security: Cameras at store entrances or self-checkout stations run vision models on-premises to detect fraud or theft in real time, without streaming sensitive video to external servers.
• AR and VR: Headsets process hand and eye tracking directly on the device for low-latency interaction. This is critical for a natural experience and avoids dependence on an internet connection.
• Smart cities: Traffic management systems process data from thousands of cameras and sensors at intersections, adjusting lights in real time to reduce congestion and idling.

What’s Involved in Inference at Scale

AI needs cloud infrastructure, edge compute, connectivity, and security. As “intelligence gets too cheap to measure” per Sam Altman, the value of the edge rises.  When the unit cost of inference collapses, you will be able to push more of it closer to the source. 

Two constraints shape edge performance:

• Capabilities. Smaller, more efficient models are improving quickly and run well on limited hardware. We’re now seeing 300M-10B capable models operating on the edge with multimodal capabilities. What was only possible on the cloud will soon be possible in a form factor such as your phone due to model improvements alone.
• Hardware and energy. The edge shares the same objective as gigawatt-scale AI data centers: achieving maximum efficiency in power and resources. The distinction lies in its limits. Instead of vast racks of servers with steady grid power and cooling, edge devices must work within small, tightly integrated systems of chips and sensors, often running on batteries with strict energy budgets.

Hybrid AI: the Edge plus the cloud

Phones, drones, robots, and vehicles will act like mobile, resource‑limited data centers. They sense locally, decide locally, and sync with the cloud to provide real-time operations and visibility, including multi-object task planning, similar to human organized activities. 

Connectivity becomes increasingly important as we start to see more autonomous systems needing to share data in real time for safety, enhanced operations, and coordination and planning. Companies like Qualcomm (QCOM), Mediatek (TPE: 2454), and Analog Devices (ADI) currently play here and and their role is expected to grow. ADI describes its role bluntly: “virtually every wireless call, text, and download today passes through an Analog Devices IC.” That spans 5G radios, lidar and radar signal chains, and more. 

Cybersecurity is the final layer of the edge inference stack. It ensures that on-device sensors, processors, and hardware operate securely and that data moving between the edge and the cloud cannot be compromised. As more intelligence shifts to distributed devices, secure operations at the edge will become as critical as cloud defenses, enabling reliable fleets of robots, vehicles, and connected systems.

Cloudflare’s (NET) “connectivity cloud” and Zero Trust stack now bridge the internet and the personal or robotic edge with global, at‑the‑edge policy enforcement. Palo Alto Networks (PANW) and CrowdStrike (CRWD) are critical here, too.

Many companies play directly in or benefit from both cloud and edge computing, while others primarily stay in single lanes. The following is an example of one that does both. 

Infineon is a ~$50B market cap company, around 80 times smaller than NVIDIA, but it plays a different role in AI. Its core businesses span automotive semiconductors, industrial power systems, sensors and microcontrollers, and secure connectivity. Together, these stacks supply the energy efficiency and reliability that advanced computing, robotics, and vehicles require.

A key focus is wide-bandgap semiconductors. Gallium Nitride (GaN) enables high-frequency, mid-voltage efficiency and is widely used in chargers, converters, and compact robotics systems. Silicon Carbide (SiC) supports very high voltage and high power, making it central to EV drivetrains, industrial motors, and AI data center power supplies.

Infineon is scaling both: it acquired GaN Systems to strengthen its GaN portfolio, is preparing 300 mm GaN wafers for production, and operates a leading SiC business with a new 200 mm fab and roughly 20 percent global share in power discretes.

Recent moves underline its overall strategy, which mirrors much of the direction of the future Robotics and AI economy. Infineon is working with NVIDIA to integrate the latest Jetson Thor robotics modules with its sensors and actuators for humanoid motion control. Infineon also acquired Marvell’s automotive Ethernet business for $2.5B, expanding into networking for autonomous and software-defined vehicles and robots. 

Despite its importance and potential, Infineon trades at 3.2x forward EV/sales and 23.7x forward P/E (FactSet consensus, September 2025), far below high-profile AI hardware peers.

Infineon is an example of just one company in ROBO and THNQ that provides the muscle and nerves of physical AI. By enabling power efficiency, sensing, and connectivity at the edge, it illustrates how convergence between data centers, robotics, and automotive systems is taking shape

Spotlight: Ambarella’s edge wins and vision

Let’s dig into another player at the focal point of edge intelligence: Ambarella (AMBA). The company is another established player with a strong foothold that is expanding into burgeoning industries. 

Ambarella began as a specialist in image signal processing for security and consumer cameras. Today, it is repositioning as a broader edge AI company, delivering system-on-chips (SoCs) that combine video, vision, and AI acceleration for robotics, automotive, and drones.

Its edge AI chips are designed for real-time computer vision. Static imaging processors remain important for cameras and surveillance. AI-accelerated vision SoCs add neural network inference, enabling autonomous navigation, driver assistance, and intelligent robotics. Together, these two product lines give Ambarella a bridge from its legacy camera base into higher-growth AI end markets.

CEO Feng Ming Wang put it simply: “It is a very exciting time for Ambarella. After a multi-year period of significant edge AI R&D investment, our broad product portfolio enables us to address a rising breadth of edge AI applications.” 

On the most recent earnings call, management announced its first design win for an on-premise AI appliance with a global networking customer, validating its edge infrastructure strategy. The company also partnered with Insta360 on a new AI-powered drone platform, strengthening its role in aerial autonomy.

Ambarella represents a niche leader moving from image capture to AI-driven perception and autonomy. For ROBO and THNQ, the company represents a convergence play between vision, inference, and robotics, showing how edge AI is expanding far beyond the data center into devices that see and act in the physical world.

Ultimately, when you operate on the edge, power, energy efficiency, and weight or form factor matter most. Companies within the ROBO Global Robotics and Automation Index (ROBO) and the ROBO Global Artificial Intelligence Index (THNQ)  sit at the intersection of physical automation and digital AI, providing the infrastructure and applications that enable autonomy to scale.

For more on robotics, AI, and healthcare, please join our upcoming webcast on October 8 at 11 a.m. ET. Register here.

Looking for regular updates? Subscribe here for weekly insights on robotics, AI, and healthcare technology, delivered straight to your inbox.

For more news, information, and analysis, visit the Disruptive Technology Content Hub.

ROBO is the underlying index for the ROBO Global Robotics & Automation ETF (ROBO), the L&G ROBO Global Robotics and Automation UCITS ETF (ROBO.LN), and the Global X ROBO Global Robotics & Automation ETF (ROBO.AU).

THNQ is the underlying index for the ROBO Global Artificial Intelligence ETF (THNQ) and the L&G Artificial Intelligence UCITS ETF (AIAI.LN).

VettaFi is the index provider for ROBO ETFs, THNQ ETF, and AIAI.LN, for which it receives an index licensing fee. However, ROBO ETFs, THNQ ETF, and AIAI.LN are not issued, sponsored, endorsed, or sold by VettaFi. VettaFi and its affiliates have no obligation or liability in connection with the issuance, administration, marketing, or trading of ROBO ETFs, THNQ ETF, and AIAI.LN.

Earn free CE credits and discover new strategies